Intelligent optimum design with the support of internet techniques

Tools

Amin, N., 2003. Intelligent optimum design with the support of internet techniques. PhD, Nottingham Trent University.

Preview

Text
10290341.pdf - Published version
Download (33MB) | Preview

Official URL: http://gateway.proquest.com/openurl?url_ver=Z39.88...

Abstract

The aim of the research is to investigate the feasibility of an efficient implementation of Internet to engineering design, with particular reference to gear design optimisation and electronic cataloguing. The research has contributed a number of originalities into this area, as listed below:

- Implementation of a large size and complicated design software package over the Internet, in a multi-user environment.

- Dynamic creation of 2-D design drawing in the form of DXF (Data Exchange Format).

- Estimation of the output of a genetic algorithm, using Artificial Neural Networks (ANN) to improve the speed of the execution of the program.

- Improvement of the learning process of the backpropagation learning algorithm.

A gear optimisation software is used as an application area and research is carried out to implement this design software over the Internet in an efficient way. This would enable the geographically dispersed companies or remote clients to have full access to the design software at any time. As one of the key issues of the system, a method to remotely execute a large size software package over the Internet has been developed.

This optimisation program utilises Genetic Algorithm (GA) to search for the best possible solution. It works in a cascade fashion, comprising of two tiers. GA is implemented in both these tiers. The output of tier one represents the initial values to tier 2, which in turn carries out a fme-tuning task. ANN is integrated into the system to estimate the output of tier 1. This improves the speed of execution of the optimisation program dramatically, since the ANN execution time is negligible.

An investigation was also carried out to improve the performance of ANN backpropagation learning algorithm. It is proposed to re-scale the output data to lie slightly below and above the two extreme values of the full range neural activation function. This allows a more weight change for the output neurons that have reached the saturation.

The final area of this research project is the creation of a prototype online design catalogue for a gear manufacturing company. A novel feature of this catalogue is its ability to dynamically create the drawings in DXF format. These drawings can then be imported into CAD software that can accept DXF format, such as AutoCAD or ProEngineer.